This invention relates generally to systems for comparing patterns, and more particularly, to a method and apparatus for comparing true and complement images of a subject pattern against true and complement images of a plurality of stored patterns, and ascertaining the one of the stored patterns which most closely matches the subject pattern. A one to one comparison can also be made to find how closely matched a subject pattern is to a reference pattern.
There is a need in the fields of pattern recognition, error correction, associative memory, computer vision, artificial intelligence systems, and other disciplines for a simple and economical system which reliably can recognize patterns. In addition to pattern recognition, such a system preferably should function as associative memory for one-dimensional and two-dimensional information.
In the optical domain, the prior art has endeavored to achieve the benefits of the invention herein using match filtering techniques. Such techniques, however, suffer from insufficient sensitivity, and are entirely inadequate for analysis of digitized patterns. In the practice of matched filtering techniques, images are transformed into Fourier space and compared against another transformed image. Spatial correlation is carried out after a search for steepest gradient allows selection of the best match.
Techniques based on neural networks have been suggested in the prior art for accomplishing associative memory tasks. These techniques, however, in addition to being cumbersome and difficult to use, suffer from the further drawback of having low storage capacity even with state-of-the-art devices. Neural network concepts have been employed in optical and optoelectronic architectures for recognition and associative memory. In the implementation of such systems, a predetermined number of stable vectors are stored using a connectivity mask. An integrating threshold feedback scheme permits an incorrect vector to modify itself to achieve similarity with a stored stable vector. Although this approach is applicable to two-dimensional patterns, the connectivity mask requires a four-dimensional system, rendering this type of system to be difficult to implement.
A number of techniques are known in the prior art for achieving pattern recognition using software. These software-based pattern matching systems suffer from the drawback that they are quite slow in their operation and have poor resolution. Moreover, software-based approaches require expensive and complex computer systems in their operation.
It would appear that the best match would be found by a direct comparison of the subject image with each of a set of stored reference images. Thus, for example, if an image on a first transparency is superimposed on its complement image on a second transparency, the resulting image would allow only minimum light to pass through both transparencies. Conversely, if an image on a transparency is placed on a copy of itself, maximum light would pass through. It is, however, a problem with this approach that if a totally blackened sheet were placed on the image, the output light would still be minimum, while in the second case, if a totally transparent sheet is placed on the test image, maximum light would pass through. Thus, the determination of whether the light passing through the transparencies is a maximum or minimum is not indicative of a match actually having been found. Thus, a spatial correlation is required if direct images are to be matched.
It is, therefore, an object of this invention to provide a simple, economical, and high speed system for comparing a subject pattern against plural stored reference patterns, and ascertaining the closest match.
It is another object of this invention to provide a system for comparing an electronically recorded subject image against a plurality of reference images stored on transparencies, and ascertaining the closest match.
It is additionally an object of this invention to determine the closeness of match between a subject image and a single reference image.
It is also an object of this invention to provide a system for comparing an electronically recorded subject image against a plurality of reference images stored electronically, and ascertaining the closest match.
It is a further object of this invention to provide a system for comparing an image stored on a transparency against a plurality of reference images stored electronically, and ascertaining the closest match.
It is additionally an object of this invention to provide a system which provides a reliable indication of a match between a plurality of patterns.
It is yet a further object of this invention to provide a system which provides a reliable indication of a match between a plurality of patterns, without need of a pixel-by-pixel computer analysis. The pixel being the smallest resolution of the image, can be arbitrarily small in this approach without reducing the capabilities of the system.
It is also another object of this invention to provide a system which provides a reliable indication of a match between a plurality of patterns which is very fast, and therefore can be employed to find a match to a subject image in a large reference pattern library.
It is yet an additional object of this invention to provide a system which provides a reliable indication of a match between a plurality of patterns without need of observation of the patterns by a human operator.
It is still another object of this invention to provide a system which permits a large number of reference images to be stored and made available quickly for comparison with the subject image.
It is a yet further object of this invention to provide a system for comparing images, wherein the comparison is achieved in an extremely short period of time.
It is also a further object of this invention to provide a system for comparing images which contain gray scales.
It is additionally another object of this invention to provide an edge detection system.
A still further object of this invention is to provide a system wherein a best match between a subject image and a one of a plurality of reference images is easily ascertained by determining a mathematical extrema.
An additional object of this invention is to provide a system wherein a best match between a subject image and a one of a plurality of reference images is easily ascertained by determining a gross transmission light level.
Yet another object of this invention is to provide a system wherein a subject image can be compared to a plurality of stored reference images simultaneously.